Facial Expression Recognition using Residual Convnet with Image Augmentations

During the COVID-19 pandemic, many offline activities are turned into online activities via video meetings to prevent the spread of the COVID 19 virus. In the online video meeting, some micro-interactions are missing when compared to direct social interactions. The use of machines to assist facial expression recognition in online video meetings is expected to increase understanding of the interactions among users. Many studies have shown that CNN-based neural networks are quite effective and accurate in image classification. In this study, some open facial expression datasets were used to train CNN-based neural networks with a total number of training data of 342,497 images. This study gets the best results using ResNet-50 architecture with Mish activation function and Accuracy Booster Plus block. This architecture is trained using the Ranger and Gradient Centralization optimization method for 60000 steps with a batch size of 256. The best results from the training result in accuracy of AffectNet validation data of 0.5972, FERPlus validation data of 0.8636, FERPlus test data of 0.8488, and RAF-DB test data of 0.8879. From this study, the proposed method outperformed plain ResNet in all test scenarios without transfer learning, and there is a potential for better performance with the pre-training model. The code is available at https://github.com/yusufrahadika-facial-expressions-essay.

Increasing The Capacity of Headstega Based on Bitwise Operation

Headstega (Head steganography) is a noiseless steganography that used email headers as a cover for concealing messages. However, it has less embedding capacity and it raises suspicion. For overcoming the problem, bitwise operation is proposed. In the proposed method, the message was embedded into the cover by converting the message and the cover into binary representation based on a mapping table that was already known by the sender and the receiver. Furthermore, XOR bitwise operations were applied to the secret message and cover bits based on random numbers that were generated using a modular function. Moreover, the result was converted into characters that represent the secret message bits. After embedding the message into the cover, an email alias was generated to camouflage the secret message characters. Finally, the sender sends the embedded cover and the email alias to the recipient. Using the proposed method, the embedding capacity is 89% larger than using the original Headstega. For reducing the adversary’s suspicion, the existing email address was used instead of creating a new email address.

Student Academic Mark Clustering Analysis and Usability Scoring on Dashboard Development Using K-Means Algorithm and System Usability Scale

Learning activities are one of the processes of delivering information or messages from teachers to students. SMPN 4 Sidoarjo is a State Junior High School (JHS) located in Sidoarjo Regency. During the learning process, the collected academic score data were still not well organized by teachers and school principals in monitoring student learning performance. The score data is from Bahasa Indonesia subject from a teacher with 222 data included at 2019/2020 school year. The method used in student clustering is K-Means. The number of clusters are determined using the elbow method and displayed in graphic form. Clustering result can be used as a reference for teachers in determining study groups and determining the best treatment for each cluster. The best clustering results are proven by validation score using Davies-Bouldin Index, Silhouette Width, and Calinski-Harabasz Index. Three clusters were obtained for each class level of data, while the cluster ranges from two to five for the data for each study group. The dashboard is used in order to visualize the clustering result. Usability testing using System Usability Scale (SUS) has a score value of 87.5, which means that the dashboard can be accepted by SMPN 4 Sidoarjo.

The Stock Exchange Prediction using Machine Learning Techniques: A Comprehensive and Systematic Literature Review

This literature review identifies and analyzes research topic trends, types of data sets, learning algorithm, methods improvements, and frameworks used in stock exchange prediction. A total of 81 studies were investigated, which were published regarding stock predictions in the period January 2015 to June 2020 which took into account the inclusion and exclusion criteria. The literature review methodology is carried out in three major phases: review planning, implementation, and report preparation, in nine steps from defining systematic review requirements to presentation of results. Estimation or regression, clustering, association, classification, and preprocessing analysis of data sets are the five main focuses revealed in the main study of stock prediction research. The classification method gets a share of 35.80% from related studies, the estimation method is 56.79%, data analytics is 4.94%, the rest is clustering and association is 1.23%. Furthermore, the use of the technical indicator data set is 74.07%, the rest are combinations of datasets. To develop a stock prediction model 48 different methods have been applied, 9 of the most widely applied methods were identified. The best method in terms of accuracy and also small error rate such as SVM, DNN, CNN, RNN, LSTM, bagging ensembles such as RF, boosting ensembles such as XGBoost, ensemble majority vote and the meta-learner approach is ensemble Stacking. Several techniques are proposed to improve prediction accuracy by combining several methods, using boosting algorithms, adding feature selection and using parameter and hyper-parameter optimization.

Brain Tumors Detection By Using Convolutional Neural Networks and Selection of Thresholds By Histogram Selection

Brain tumors in medical images have a high diversity in terms of shape and size. Some of the data found a form between the tumor tissue and normal tissue, whereas knowing the tumor’s profile and characteristics becomes a crucial part of searching. By using machine learning capabilities, where machines are given several variables and provide decisions to a certain degree, they have broadly given decisions that support subject matter in making decisions. This study applies the threshold selection method using histogram selection on CT scan data, while the appropriate threshold selection method selects the tumor position accordingly. Furthermore, the Convolutional Neural Network (CNN) is used to classify whether the selected image is a tumor or not. Using CT scan data and calculated experiments, this algorithm can finally be approved and given a brain classification with an accuracy of 75.42 percent.

Comparison of FairMOT-VGG16 and MCMOT Implementation for Multi-Object Tracking and Gender Detection on Mall CCTV

The crowd detection system on CCTV has proven to be useful for retail and shopping sector owners in mall areas. The data can be used as a guide by shopping center owners to find out the number of visitors who enter at a certain time. However, such information was still insufficient. The need for richer data has led to the development of more specific person detection which involves gender. Gender detection can provide specific information on the number of men and women visiting a particular location. However, gender detection alone does not provide an identity label for every detection that occurs, so it needs to be combined with a multi-person tracking system. This study compares two tracking methods with gender detection, namely FairMOT with gender classification and MCMOT. The first method produces MOTA, MOTP, IDS, and FPS of 78.56, 79.57, 19, and 24.4, while the second method produces 69.84, 81.94, 147, and 30.5. In addition, evaluation of gender was also carried out where the first method resulted in a gender accuracy of 65\% while the second method was 62.35\%.

COVIWD: COVID-19 Wikidata Dashboard

COVID-19 (short for coronavirus disease 2019) is an emerging infectious disease that has had a tremendous impact on our daily lives. Globally, there have been over 95 million cases of COVID-19 and 2 million deaths across 191 countries and regions. The rapid spread and severity of COVID-19 call for a monitoring dashboard that can be developed quickly in an adaptable manner. Wikidata is a free, collaborative knowledge graph, collecting structured data about various themes, including that of COVID-19. We present COVIWD, a COVID-19 Wikidata dashboard, which provides a one-stop information/visualization service for topics related to COVID-19, ranging from symptoms and risk factors to comparison of cases and deaths among countries. The dashboard is one of the first that leverages open knowledge graph technologies, namely, RDF (for data modeling) and SPARQL (for querying), to give a live, concise snapshot of the COVID-19 pandemic. The use of both RDF and SPARQL enables rapid and flexible application development. COVIWD is available at http://coviwd.org.

Reducing Adversarial Vulnerability through Adaptive Training Batch Size

Neural networks possess an ability to generalize well to data distribution, to an extent that they are capable of fitting to a randomly labeled data. But they are also known to be extremely sensitive to adversarial examples. Batch Normalization (BatchNorm), very commonly part of deep learning architecture, has been found to increase adversarial vulnerability. Fixup Initialization (Fixup Init) has been shown as an alternative to BatchNorm, which can considerably strengthen the networks against adversarial examples. This robustness can be improved further by employing smaller batch size in training. The latter, however, comes with a tradeoff in the form of a significant increase of training time (up to ten times longer when reducing batch size from the default 128 to 8 for ResNet-56). In this paper, we propose a workaround to this problem by starting the training with a small batch size and gradually increase it to larger ones during training. We empirically show that our proposal can still improve adversarial robustness (up to 5.73\%) of ResNet-56 with Fixup Init and default batch size of 128. At the same time, our proposal keeps the training time considerably shorter (only 4 times longer, instead of 10 times).

Sarcasm Detection Engine for Twitter Sentiment Analysis using Textual and Emoji Feature

Twitter is a social media platform that is used to express sentiments about events, topics, individuals, and groups. Sentiments in Tweets can be classified as positive or negative expressions. However, in sentiment, there is an expression that is actually the opposite of what is mean to be, and this is called sarcasm. The existence of sarcasm in a Tweet is difficult to detect automatically by a system even by humans. In this research, we propose a weighting scheme based on inconsistency between sentimen of tweet contain in Indonesian and the usage of emoji. With the weighting scheme for the detection of sarcasm, it can be used to find out a sentiment about a event, topic, individual, group, or product's review. The proposed method is by calculating the distance between the textual feature polarity score obtained from the Convolutional Neural Network and the emoji polarity score in a Tweet. This method is used to find the boundary value between Tweets that contain sarcasm or not. The experimental results of the model developed, obtained f1-score 87.5%, precision 90.5% and recall 84.8%. By using the textual features and emoji models, it can detect sarcasm in a Tweet.

Automatic Image Self-Enhancement for Multi-Scale Spectral on Low Resolution Video

MSR technique is a process used to reduce the search area in an image. MSR relies heavily on image salience from image capture devices. The purpose of this study is to improve the detection of people's objects on video by increasing the quality of the image frames on video and MSR. This research uses artificial video taken in a room installed by CCTV. Object detection is evaluated using precision, recall, and fscore values. The results showed an increase in the quality of the performance of object detection that was properly detected as a person. The average detection performance is indicated by an fscore of 14.47%. Increasing the quantity of detection of objects of people reached 168.8% compared to the detection of only using MSR